Context Engineering Emerges as Enterprise AI Agent Battleground
36氪 人工智能 reports that a telecom subsidiary used an Agent platform to reduce monthly invoice-data reconciliation from 31 people working for two days to one worker taking two hours, though the workflow covered only part of the process. The analysis says context engineering is becoming central to enterprise agents, while noting unresolved barriers involving system integration, permissions, latency, cost, organizational silos, and security. It also cites iResearch data indicating broad AI adoption but limited progression into workflow optimization and specialized business processes.
The reported use case shows measurable efficiency gains from an enterprise Agent, but the incomplete workflow and unresolved integration, governance, and cost issues illustrate what must be addressed before agents can operate broader business processes.
“31个人干2天的活，现在1个人2小时就能搞定。”
这是某头部电信运营商子公司引入AI Agent后，一项财务工作的效率变化。过去，31个省的财务人员每月要花两天，核对打款、订单和开票需求三张表；如今，通过Agent平台，企业将匹配规则沉淀为Skill，一名员工只需两小时，就能完成数万张发票的数据整理。
但在该公司副总经理看来，这还算不上真正意义上的业务全流程智能化。“我们现在只是完成了中间一段。”他直言，企业真正想要的是从接收需求、开票到发送电子发票的“端到端”自动流转。
这盆冷水，恰恰浇出了当下企业级Agent落地的真实温差。
一边是底层大模型的技术狂飙，从复杂推理、自主规划到多智能体协作，AI看起来无所不能；另一边，当这些能力真正撞上企业的既有业务体系，却步履维艰。
艾瑞咨询的一组数据，撕开了这种“浅层繁荣”的面纱。
#Anthropic#Feishu#Harvey#LangChain#Redis#Coze#Dify#DeepSeek#McKinsey#iResearch